What is game theory… Game theory studies settings where multiple parties (agents) each have –different preferences (utility functions), –different actions that they can take Each agent’s utility (potentially) depends on all agents’ actions –What is optimal for one agent depends on what other agents do Very circular! Game theory studies how agents can rationally form beliefs over what other agents will do, and (hence) how agents should act –Useful for acting as well as predicting behavior of others

Example: Suppose we have the following game: each person submits a number between 0 and 100. The winner is the person whose number is 2/3’s of the average number submitted. What will you submit? Let’s try it.

Soccer Penalty kick example probability.7 probability.3 probability.6 probability.4 select direction of kick with some probability Does the goalie want to change its probabilities? action

Back to regular class notes… Extensive form games Each level in the tree corresponds to one player. They take turns. Not simultaneous play as in normal form games. Rewards are listed at the leaves. Can model normal form by making nodes indistinguishable

Game playing & AI perfect information games: no uncertainty about the state of the game (e.g. tic- tac-toe, chess, Go) imperfect information games: uncertainty about the state of the game (e.g. poker) 1 gets King1 gets Jack bet stay callfoldcallfoldcallfoldcallfold “nature” player 1 player 2 white black Qa1-a8Qa1-f6 Kf8-f7Kf8-g7Kf8-g8Kf8-e8 black winswhite winsdraw Optimal play: value of each node = value of optimal child for current player (backward induction, minimax) For chess and Go, tree is too large –Use other techniques (heuristics, limited-depth search, alpha-beta, pruning, …) Top computer programs (arguably) better than humans in chess, not yet in Go Player 2 cannot distinguish nodes connected by dotted lines –Backward induction fails; need more sophisticated game-theoretic techniques for optimal play Small poker variants can be solved optimally Humans still better than top computer programs at full-scale poker (at least most versions) Top computer (heads-up) poker players are based on techniques for game theory

Questions and problems in (computational) game theory How should we represent games (=strategic settings)? –Standard game-theoretic representations not always concise enough What does it mean to solve a game? –Solution concepts from game theory, e.g., Nash equilibrium –Nash equilibrium: if the other player doesn’t switch, you don’t want to either. 100,1000,0 100,100

How computationally hard is it to solve games? –Can we solve them approximately? Is there a role for (machine) learning in games? What types of modeling problems do we face when addressing real-world games? –E.g., applications in security …

What is social choice? “Social choice theory is a theoretical framework for measuring individual interests, values, or welfares as an aggregate towards collective decision.” [Wikipedia, Aug. 2009] I.e., making decisions based on the preferences of multiple agents Largely, but not exclusively, focused on voting

Voting over outcomes > > > > voting rule (mechanism) determines winner based on votes. Can vote over other things too –Where to go for dinner tonight, other joint plans, … Many different rules exist for selecting the winner

Combinatorial auctions v( ) = $500 v( ) = $700 v( ) = $300 Simultaneously for sale:,, bid 1 bid 2 bid 3 used in truckload transportation, industrial procurement, radio spectrum allocation, …

Kidney exchange Donor isn’t willing to donate unless friend gets a kidney. patient 1 donor 1 (patient 1’s friend) patient 2 donor 2 (patient 2’s friend) patient 3 donor 3 (patient 3’s friend) patient 4 donor 4 (patient 4’s friend) compatibilities

Problems in computational social choice Winner determination problem –For some voting rules, determining the winner is NP-hard –In a combinatorial auction, deciding which bids win is (in general) an NP-hard problem Preference elicitation (communication) problem –Can be impractical to communicate all of one’s preferences (e.g., valuation for every bundle) Mechanism design problem –How do we get the bidders to behave so that we get good outcomes? We don’t want to reward lying. These problems interact in nontrivial ways –E.g. limited computational or communication capacity can limit mechanism design options –… but can perhaps also be used in a positive way

What is mechanism design? Assume –that a game “designer” chooses the game structure rather than inheriting one –that the designer is interested in the game’s outcome usually solved by motivating agents to disclose their private information.” [Wikipedia, Aug. 2010]

Mechanism design… Mechanism = rules of auction, exchange, … A function that takes reported preferences (bids) as input, and produces outcome (allocation, payments to be made) as output $ 750 $ 450 $ 800 v( ) = $400 v() = $600 $ 400 v( ) = $500 v() = $400 f()= The entire function f is one mechanism On example: the mechanism from before: find allocation that maximizes (reported) utilities, distribute (reported) gains evenly Other mechanisms choose different allocations, payments

Example of mechanisms: (single-item) auctions Sealed-bid auction: every bidder submits bid in a sealed envelope First-price sealed-bid auction: highest bid wins, pays amount of own bid Second-price sealed-bid auction: highest bid wins, pays amount of second-highest bid 0 bid 1: $10 bid 2: $8 bid 3: $1 first-price: bid 1 wins, pays $10 second-price: bid 1 wins, pays $8

Which auction generates more revenue? Each bid depends on –bidder’s true valuation for the item (utility = valuation - payment), –bidder’s beliefs over what others will bid (→ game theory), –and... the auction mechanism used In a first-price auction, it does not make sense to bid your true valuation –Even if you win, your utility will be 0… In a second-price auction, (we will see later that) it always makes sense to bid your true valuation 0 bid 1: $10 bid 2: $8 bid 3: $1 Are there other auctions that perform better? How do we know when we have found the best one? 0 bid 1: $5 bid 2: $4 bid 3: $1 a likely outcome for the first-price mechanism a likely outcome for the second- price mechanism

Mechanism design… Mechanism = game We can use game theory to predict what will happen under a mechanism –if agents act strategically When is a mechanism “good”? –Should it result in outcomes that are good for the reported preferences, or for the true preferences? –Should agents ever end up lying about their preferences (in the game-theoretic solution)? –Should it always generate the best allocation – the one who wants it the most gets it? –Should agents ever leave money on the table? Can we solve for the optimal mechanism?

Many uses of linear programming, mixed integer (linear) programming in this course Linear programmingMixed integer linear programming Game theory Dominated strategies Minimax strategies Correlated equilibrium Optimal mixed strategies to commit to Nash equilibrium Optimal mixed strategies to commit to in more complex settings Social choice, expressive marketplaces Winner determination in auctions, exchanges, … with partially acceptable bids Winner determination in: auctions, exchanges, … without partially acceptable bids; Kemeny, Slater, other voting rules; kidney exchange Mechanism design Automatically designing optimal mechanisms that use randomization Automatically designing optimal mechanisms that do not use randomization

Explain the game theory issues behind sponsored search. Explain the game theory issues behind charitable giving Explain the concept of a prediction market How could the overall outcome be worse if you made decisions on information which doesn’t correspond to true valuation?

NP hardness Since NP hardness is a worst case measure, it may be easy to find an effective way of misreporting one’s preferences most of the time. What is the issue here?

Computer Science and Game Theory PPAD – new complexity class to handle problems for which a solution is known to exist. How can a solution be known to exist if we haven’t produced it? Give an example of the price of anarchy revelation/hiding What is bounded rationality? Ex. maxmin or NE strategy not typical in solutions BDI (“beady eye”) architecture